應用資料探勘技術於股票投資-以鴻海為例

Applications of Data Mining Techniques for Stock Investment – A Case Study of Foxconn

陳均姿(Chun-Tzu Chen)

指導教授 ： 巫木誠 ； 洪暉智

國立交通大學/管理學院/工業工程與管理系所/碩士(2018年)

摘要

本研究應用資料探勘技術於股票投資，並以鴻海為研究標的。買點信號定義為未來90內，有一天(含)以上之收盤價漲幅超過10%。而輸入變數為25個可能影響股價變動之因素；輸出變數則為是否為買點信號。本研究之資料研究期間為2008年1月至2017年12月共十年之資料。交易方法則分為三個階段。第一階段為利用原始資料的65%為訓練資料集，並以不同的資料探勘技術建立買點信號預測模型。但因買點信號預測模型之準確度能力不足，因此，本研究提出保守的交易決策，依買點信號出現一次、連續出現兩次或更多次為買點，進行投資。第二階段為利用原始資料的10%為驗證資料集，選出最佳的買點信號預測模型和交易決策。第三階段則將所選的最佳買點信號預測模型和交易決策，以原始資料集的後25%之測試資料集進行測試。從實驗結果可得，所選的交易方法之投資報酬率為0.18%，優於基準(Benchmark)之投資報酬率-2.33%。

關鍵詞：資料探勘、交易決策、投資報酬率

關鍵字

資料探勘 ； 交易決策 ； 投資報酬率

並列摘要

This research applies a stock trading methodology based on data mining (DM) techniques to the case of Foxconn Inc. A trading day is defined as a “Buy-Signal” if the stock price shall rise over 10% in the coming trading period (say, 90 days). The input of a DM technique involves 25 variables that might affect stock price; and the output is a binary variable (Buy or Not-Buy). This study involves ten years of given input/output data (2008/1-2017/12). The trading methodology involves three phases. Firstly, B-signal predictors are trained using different DM algorithms based on the first 65% of data (training dataset). Trained B-signal predictors are far from perfect in prediction accuracy; and one might propose conservative trading policies that take buy-action only when two or more (m) consecutive B-signal appears. Secondly, the best combination of B-signal predictor and trading policy is selected based on the subsequent 10% of data (validation dataset). Thirdly, the selected combination of B-signal predictor and trading policy is tested using the last 25% of data (testing dataset). Numerical experiments reveal that the return on investment of the selected trading method is 0.18%, which outperforms that -2.33% of the benchmark method.
Keyword: Data Mining, Trading Policy, Return On Investment

並列關鍵字

Data Mining ； Trading Policy ； Return On Investment

參考文獻

Breiman, L. (2001). Random forests. Machine learning, 45(1), 5-32.

Breiman, Leo, Jerome Friedman, R. Olshen and C. Stone (1984). Classification and Regression Trees. Belmont, California: Wadsworth.

Chong, E., Han, C., & Park, F. C. (2017). Deep learning networks for stock market analysis and prediction: Methodology, data representations, and case studies. Expert Systems with Applications, 83, 187-205.

Cortes, C., & Vapnik, V. (1995). Support-vector networks. Machine learning, 20(3), 273-297.

Cover, T., & Hart, P. (1967). Nearest neighbor pattern classification. IEEE transactions on information theory, 13(1), 21-27.